OpenAI Agents API
Points and comments are a snapshot, not live.
OpenAI launches a managed Agents API for building and deploying autonomous AI agents.
The Agents API provides tools to create, configure, and run persistent agent sessions with built-in sandbox environments (OpenAI-hosted or self-hosted), web search, function calling, MCP integrations, and multi-agent orchestration. It offers managed session lifecycle, webhooks, and observability tracing. Pricing includes per-20-minute environment costs and tool usage fees. The API is distinct from the Agents SDK, which runs locally.
What commenters are saying
Commenters debate the API vs. SDK tradeoffs: the API offers ease of setup and managed security but increases vendor lock-in and relies on trusting OpenAI's sandbox enforcement. Several question its value over local SDKs or self-hosted VMs. Others note the API is for enterprise deployment, while the SDK suits development. A critical thread discusses limitations in sandbox network restriction bypass attempts. Some commenters contrast this with existing open-source alternatives.